The Next Byte Podcast · 16 Nov 2021 · 26m
Hosts Daniel and Farbod discuss how computer vision AI is applied in manufacturing and agriculture, featuring insights from Plain Sight AI and a project by the Technical University of Denmark. They explain Plain Sight AI's three levels of vision systems for detecting hazardous objects on production lines, emphasizing that Level 3 instance segmentation offers superior sorting and counting capabilities. The hosts also detail a robot developed by PhD student Ronja Goldring to autonomously remove weeds from organic dairy farms using laser or electrocution, highlighting the importance of large, diverse training datasets and collaboration with end-users.
Where this conversation describes AI being used to build or deliver a product. Part of the AI in practice tracker.
ManufacturingMouser Electronicscost less than $300
Mouser Electronics built a computer vision system to inspect packaging labels in their warehouse, ensuring logos are applied correctly and are scannable.
“They use in their, I think, in their warehouse, my bad. Before things ship out, they use this exact procedure and using this exact hardware, remember less than $300.”
Speaker 1 · Hear it at 15:14
Engineering & simulationEdge Impulse
The team used Edge Impulse to compile a neural network model for a computer vision system that detects packaging defects.
“Then you upload it to this website called Edge Impulse, which is something that a lot of people use for computer vision, creating their algorithm. You compile the model on edge impulse, install it to this machine and you can start testing right away.”
Speaker 1 · Hear it at 13:57
Engineering & simulationPlainsight AI2 to 3 times as much sorting volume
Plainsight AI developed computer vision algorithms for anomaly detection, classification, and instance segmentation to identify hazardous objects and defects on manufacturing lines.
“they say at plain sight, a CV, computer vision, AI system can typically handle 2 to 3 times as much sorting volume in a given time as a team of humans. And once it gets to level three, it is significantly more accurate as well.”
Speaker 1 · Hear it at 10:09
Engineering & simulationTechnical University of Denmark / Ronia Golden Ringone centimeter accuracy
Researchers created a large public dataset of weeds and grass images to train an algorithm for a robot that autonomously identifies and kills weeds in organic fields.
“with the Galileo system, they're able to localized where the weeds are within a one centimeter accuracy. So the robot can find it and kill it, but having that data of where they're growing would be useful to the farmers to maybe see patterns or something.”
Speaker 2 · Hear it at 22:37
early-stage companies
Other shows on the same ground, closest first.
Chris Padwick, Director of Computer Vision and Machine Learning at Blue River Technology (a John Deere subsidiary), explains the engineering behind the See & Spray system, which uses computer vision to spot-spray herbicide on weeds. He details the massive compute requirements, noting the system uses NVIDIA Xavier and Orin GPUs to achieve 0.3 petaflops of processing power for real-time image classification at 12 mph. Padwick also discusses the critical role of data collection, the acquisition by John Deere for scaling, and the future of autonomous agricultural machinery.
Willy Pell, VP of Autonomy at Blue River Technology (acquired by John Deere), explains how the company applied machine learning and computer vision to automate farming, starting with the See & Spray weed management system and expanding to the autonomous John Deere AR tractor. Pell details the strategic decision to use stereo cameras instead of LIDAR due to dust and speed constraints, and describes how the system uses deep learning to detect crops versus weeds, achieving significant herbicide savings. The episode also covers the broader roadmap for autonomous tillage, planting, and grain cart coordination, drawing inspiration from Tesla's data-driven approach to achieve full autonomy in a semi-structured agricultural environment.
Dr Petra Bosilj of the University of Lincoln explains that robotic vision in agriculture is bottlenecked by a lack of annotated training data for complex tasks like distinguishing weeds from crops, though precision spraying can reduce herbicide use by 80%. Chris Chavasse, founder of Muddy Machines, describes his startup's development of lightweight, swarm-capable field robots for harvesting tender stem broccoli, designed to reduce soil compaction compared to heavy industrial machinery. The episode contrasts academic research into phenotyping and soil health with the commercial challenges of building rugged, soft-robotic end effectors and securing funding during the pandemic.
Ken Goldberg and Jim Liefer of Ambi Robotics discuss their Sim2Real AI approach to robotic grasping, explaining how they trained neural networks on simulation data to handle diverse e-commerce objects without human intervention. Paul Mikesell, founder of Carbon Robotics, details the development of their third-generation autonomous weeding robot, which uses high-powered lasers and computer vision to identify and destroy weeds in row crops. The episode also covers WTWH Media's acquisition of Mobile Robot Guide and updates on the Mars 2020 Ingenuity helicopter.
Jorge Heraud, VP of Autonomy at John Deere, explains how the company leverages its acquisition of Blue River Technology and Spark AI to deploy Level 4 autonomous tractors, using computer vision and machine learning to enable 'See and Spray' and autonomous tillage while managing rural connectivity challenges. RJ He and Ryan McMichael, technology leaders at Zoox, discuss their purpose-built, steering-wheel-free robotaxi recently deployed on public roads in California, detailing a perception stack that integrates cameras, LIDAR, radar, and thermal imaging to ensure safety through redundancy and diversity. The episode contrasts the agricultural application of autonomy, focused on labor relief and precision, with the urban mobility focus of Zoox, highlighting the technical hurdles of sensor fusion, edge cases, and the critical importance of building a robust safety case for public acceptance.
The summary, chapters and topics are generated from the episode’s audio by this directory, not written by The Next Byte Podcast. The episode itself belongs to its publisher.